How to AI. The spine every other tool plugs into.
Most GTM tools are just connection points where the AI pulls data. The core AI tools are the foundation underneath all of them. Learn the spine first: set up Claude and ChatGPT, prompt the way that actually works in 2026, connect your tools, and build. Then go wire up your real stack.
You do not have to start at the beginning. Find yourself below and jump in.
AI rewards domain expertise more than tenure with the tools. Your years in GTM are the edge, not a handicap.
Analysis paralysis is the number one killer. Pick one section, do its exercise today, on your real work.
A search engine you query and close. A coworker you onboard once, and it compounds every day after.
Fifteen minutes on a real task beats a weekend course binge. Reps on your work, not textbook examples.
How to think about AI
YOU LEAVE WITHA mental model that stops you wasting the tool.
The gap between "I use ChatGPT sometimes" and "AI saves me ten hours a week" is almost entirely one shift: stop treating it as a search engine, start treating it as a coworker who needs onboarding.
Two rules before you touch any tool. First, workflow first: map the task on paper. What do you do repeatedly? Which parts need your judgment, and which do not? AI takes the second pile. Second, the Wrapper Test: before you pay for any "AI for X" tool, ask whether Claude plus a good prompt and your own context already does it. Most of the time, yes. Buy the plumbing, build the edge.
I'm a [your role] in GTM. Here's a task I do manually every week: [describe it in 3-4 sentences]. Act as an operator, not a cheerleader. Tell me: which parts of this a good AI setup could take off my plate, which parts still need my judgment, and the single highest-leverage place to start. No hype.
The tools that look most impressive are often a thin wrapper on a model you already pay for. Run the Wrapper Test before every purchase.
Set up your AI (Claude & ChatGPT)
YOU LEAVE WITHAn AI that already knows your context before you ask.
This is Level 1, and almost everyone skips it, which is exactly why their results are mediocre. Do not work in the raw chat window.
Set up a Project (Claude, ChatGPT, and Gemini all have them). Inside it, add three things: custom instructions (persistent rules it follows every time), reference docs (your ICP, positioning, brand voice, one sample of your best work), and memory (facts it keeps across sessions). Turn web search on so it stops answering from stale training data.
This is the onboarding. You do it once, and every prompt after is smarter because the context is already loaded.
You are my GTM operating partner. Context: we sell [product] to [ICP]. Positioning: [one line]. My role: [role]. Rules: be direct and concise, no preamble, no flattery. When you need information I haven't given you, ask before assuming. Default to specifics and examples. When you make a claim about our numbers or market, flag what you're inferring vs. what I actually told you.
The difference between "AI is a toy" and "AI is my best hire" is almost entirely whether you set up Projects and context first. Level 1, and most people never do it.
Prompt + context (the 2026 truth)
YOU LEAVE WITHReliably good output, using the rules that actually work now.
Start simple: Persona + Task + Context + Format. "You are a senior RevOps analyst. Turn this pipeline export into a five-bullet risk summary for my VP. Data: [paste]. Lead with the biggest risk." That covers most of it.
But here is what nobody updated you on: a lot of the prompt-engineering advice from 2024 and 2025 now actively hurts your results. The real skill in 2026 is context engineering. The frame that stuck (from Andrej Karpathy): the model is a CPU, the context window is its RAM, and your job is to be the operating system that loads exactly the right information for the task. What you load matters more than how cleverly you phrase the ask.
- "Think step by step" now hurts modern reasoning models. They reason internally, so the instruction is redundant or harmful.
- ALL-CAPS, "YOU MUST," and "NEVER EVER" make Claude worse, not better. Calm and direct wins.
- Long prompts degrade. Reasoning falls off around 3,000 tokens; the sweet spot is 150 to 300 words. Load context as attached files, not a wall of text.
- Few-shot examples no longer improve reasoning. They only align format.
- For Claude, structure with XML tags, not markdown or numbered lists.
- The model is a yes-man by default. It agrees with you unless you force it not to.
I'm going to share [a decision / a draft / a plan]. Be my devil's advocate. Rules: 1) Do not validate it first. Skip the compliments. 2) Give the 3 strongest arguments against it. 3) Name the assumption I'm most likely wrong about. 4) Tell me what someone who disagrees would say, and why they might be right. 5) Only after all that, tell me what's genuinely strong. Here it is: [paste].
I shipped worse work for months using 2024 prompt tricks that quietly hurt modern models. The fix was less clever phrasing and more of the right context loaded up front.
Connect your tools inside Claude (MCP)
YOU LEAVE WITHAI that pulls from your real tools instead of what you paste.
MCP (Model Context Protocol) is the standard way to plug a tool into your AI. Think of it as a USB-C port for AI: one standard connector, and suddenly Claude can read your CRM, your calendar, your docs, your warehouse. It shipped in late 2024, exploded in 2025, and is now the default plug.
You do not build anything. In Claude, open the Connectors menu and add a tool, or paste a tool's remote MCP server URL as a custom connector and authorize with one click. In ChatGPT, it's Settings then Apps. Most connections use the tool's own login and are read-only, so you approve access and can disconnect anytime. Start with one tool you already live in.
You now have access to [tool] via connector. Before you touch anything, tell me what you can read and what you can write. Then, read-only: pull [the specific thing] and summarize it. Do not write or change anything until I say so.
Read-only first, always. Confirm what an AI can see and change before you let it act. That one habit saves you from every scary story.
Which of your tools have an MCP, and exactly how to connect each one, lives in the connection dictionary.
Open The GTM Stack →Build it (Lovable, Make, Vercel)
YOU LEAVE WITHShip a working thing without writing code.
"Vibe coding" means you describe what you want in plain English and the AI builds it. There is a spectrum. Point-and-click app builders (Lovable, Bolt, v0) are where a non-developer starts: describe an app, get a working full-stack build, deploy in a click. Make is the visual automation layer: connect apps and let workflows run on triggers. Vercel is where the thing you built goes live, hosted at a real URL.
The order that works: prove the idea in Claude first (the Wrapper Test again). If you need a real interface or a scheduled workflow, then reach for the builder. Do not start by paying for a builder. Start by proving the idea in chat.
I want to build [describe the thing and who uses it]. Act as a pragmatic builder. Before any code: 1) Tell me if I even need a custom build, or if Claude plus a connector already does this. 2) If I do, give me the simplest version that delivers value, and what to leave out of v1. 3) Recommend the right tool (Lovable, Make, or Vercel) and why. Keep it to what a non-developer can actually ship.
Half the things you'd pay a developer for, you can ship yourself in an afternoon now. The other half you shouldn't build at all, which is what the scoping prompt is for.
Make it repeat (Skills, Automations, Agents)
YOU LEAVE WITHThe work runs without you in the loop.
This is where "how to AI" becomes an operating system. Three rungs. Skills: once you've gone back and forth enough to nail a task, package that into a reusable skill you can call anytime instead of re-prompting from scratch. Automations: schedule your skills for recurring work. Agents: give it a goal instead of a script, and it reasons, picks tools, acts, checks, and loops until the job is done.
The jump from automation to agent is who decides what runs. At the automation rung, you decide what and when. At the agent rung, it does. Most people never get past prompting. The compounding gains live here.
Reverse-engineer this conversation into a reusable skill I can call anytime. Give me a clean, self-contained definition: what it does, what input it needs, and the steps, so I can save it and reuse it without re-explaining. Then tell me one recurring task where I should schedule it.
This is the operator system: solve the problem once, package it, then let it run. It is the whole point of learning the spine.
Built GTM's skills and playbooks are exactly this, packaged and ready to install.
Browse the Tools →Go deeper on the tools that matter.
Full guides on how to use, set up, and build with a specific tool. More shipping as I write them.
Do I need to know how to code?
No. Everything here works by describing what you want in plain English. Claude, Lovable, and Make write any code for you. Coding only matters if you want to train your own models, which is a different path.
Is prompt engineering still a thing in 2026?
Yes, but it changed. The lever moved from clever phrasing to context: what information you load before you ask. Several 2024 tricks, like "think step by step," now hurt modern models.
Claude or ChatGPT?
Both are excellent and the skills transfer. This guide leans on Claude because Projects and MCP connections create a clean progression, but everything applies to both.
What is MCP?
Model Context Protocol, the standard way to connect a tool to your AI so it can read and act on your real data. Think USB-C for AI. You connect it, you don't build it.
Do I need to pay for tools?
Start free. A browser and a free Claude account cover the first two sections. Run the Wrapper Test before paying for anything.
Where do I connect my specific GTM tools?
In the GTM Stack connection dictionary, which lists each provider's MCP and API and how to connect it.
You've got the spine. Now wire up your stack.
The core AI tools are the foundation. Every other tool is a connection point. The GTM Stack shows you every provider worth knowing, and how to connect its MCP or API to your AI.